R ENDICONTI
del
S EMINARIO M ATEMATICO
della
U NIVERSITÀ DI P ADOVA
B ENEDETTO S CIMEMI
Homogeneous verbal functions on groups
Rendiconti del Seminario Matematico della Università di Padova, tome 49 (1973), p. 299-321
<http://www.numdam.org/item?id=RSMUP_1973__49__299_0>
© Rendiconti del Seminario Matematico della Università di Padova, 1973, tous droits réservés.
L’accès aux archives de la revue « Rendiconti del Seminario Matematico della Università di Padova » (http://rendiconti.math.unipd.it/) implique l’accord avec les conditions générales d’utilisation (http://www.numdam.org/conditions).
Toute utilisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
BENEDETTO SCIMEMI
(*)
1. - This research was
originated by
ourprevious
paper[4],
whosemain results will be
shortly
recalled: let p be aprime number,
P afinite p-group of
exponent p-, d
a divisor ofp -1, a
aprimitive
d-th root of
unity,
mod anautomorphism
of P of orderd;
thenevery element h E P can be
uniquely
written asThe set
P1=
need not be asubgroup
of P(we
could takeany other
P; .
=(h, ;
to avoidtrivialities),
but aloop operation
o can be introduced onPl by letting
for all x, y EP,,:
x o y = If this element is
expanded
in terms of basic commutatorsin x, y
then a recursive formula
naturally yields
for theegponents eu
somerational
(non-integer) values;
allexamples
show evidence that thesenumbers are
independent
of x, y,P, q
andonly depend
on d.The motivation for the
present
research was to prove this inde-pendance,
thusshowing
a more naturalorigin
for theloop
structureswe studied in
[4]. After
we noticed that we could assume without loss ~’ =(Pl),
it seemed natural to look for a group F**admitting
(*) Indirizzo dell’A.: Seminario Matematico dell’Universith - Via Bel.
zoni 3, 35100 Padova.
Lavoro eseguito nell’ambito dei
Gruppi
di Ricerca Matematica del C.N.R.an
automorphism ø
such that: if P is any finite p-group with n gen-erators, and 99
is anautomorphism
of Pinducing
the a-th power on thesegenerators,
then P is ahomomorphic image
of F**and 99
isinduced
by
0.By using
a recent theorem of R. H. Bruck[1],
asatisfactory
answerwas found. The main theorem of the
present
paper(theor. 1,
par.5)
states the
following:
Let F be a free group of finite
rank n, generated by
the free gen- eratorsX, Y,
.... For eachpositive
number d there exist groups F*and .F’** with the
following properties:
1) .F~.F*~h**.
2)
The elements of F* may beuniquely represented
as infiniteordered
products
in terms of basic commutators inX, Y, ...
where the
exponents e.
lie in thering Z[d]
of those rational numbersrls
such that noprime dividing s
is =1 mod d.3)
The element of ~** aresimilar, with e.
in thering
obtained from
by adjunction
of aprimitive (complex)
d-th rootof
unity
a .4)
In 1** the «a-th-power » g"
is defined(for each g
EF**)
andthere exists an
automorphism 0
of ~’** such that5) Every
elementg E F* (in particular,
every word in~Y, Y, ...)
can be
uniquely
written asNow let p be a
prime number,
P a finite p-group ofexponent
pm with ngenerators x, y,
..., d a divisor of aprimitive
d-th rootof
unity,
mod pm. Then there exists aunique homomorphism
v of F**onto P such that
If q
is anautomorphism
of P such that X9, = xa,ygl
= ya, ... it is clearthat 0153fP == xa =
(Xv)a
=(Xa)v
=(X~~v,
etc. Hence it iseasily
seen thatfor every = gv we as we wanted.
The consequences of
5)
are derived in terms of « verbal functions ».A verbal function on a group is the
analog
of apolynomial
functionon a
ring.
Let H be a group, I’ the free group with freegenerators Xl, X2, ..., Xn;
to each wordf = f (X1, X2, ..., Xn) E .~’
we associatethe « verbal function in n variables
say 7, mapping
then-ple ( h1, h2 ,
... ,hn )
of elements into the elementh2 , ... , hn ) E H
which is obtained from
f by replacing
each Xby hi
andtaking
inversesand
products
in H. For n ==1,
a verbal function issimply
a power;for n = 2 among the verbal functions
(or
verbaloperations)
we findthe usual
group-multiplication
andconjugation,
associated to the wordsXl X2
and I etc.Clearly,
the verbal functions in n variables on H form agroup -P (with respect
tomultiplication
ofimages),
an
image
of F under thehomomorphism
If H is a finitep-group,
p ---1 mod d,
then there is a natural way to extend thehomomorphism f - f
to the group P**. If weapply
thishomomorphism
to the word g, factorized as in
5) above,
we derive thefollowing (Cor. 1,
par.7):
Let P be a finite p-group of exponent
pm, d
a divisor ofp -1,
a a
primitive
d-th root ofunity, mod pm.
Then every verbal func-tion g
on P can beuniquely
written as withBy analogy
withAnalysis,
a verbal functionsatisfying
the abovecondition
for g;
is said to be «a-homogeneous
ofdegree j
». Since(but
ingeneral g; ~ I’),
thehomogeneous factor g;
is obtained from a basic commutatorexpansion
which isuniquely
definedby
the word g and the
number d, through
a substitution of elements of P.If we
apply
this to the groupmultiplication
onP,
thusletting
we find for
gl
aloop-operation
on P(Theor. 2,
par. 7). In
particular,
for d =2,
wefind,
withinisomorphisms,
theloop operation
which has beenextensively
studiedby
G. Glauberman(see
ref. in[4]).
When P admits anautomorphism
of orderd,
thisoperation
iseasily proved
to induce on the «eigensubset >> Pi
the samemultiplication
we definedabove,
i.e.Apart
fromloops,
the wholesubject
in thepresent
broader contextis
strictly
connected with the work of M. Lazard[3]
onnilpotent
groups and
Lie-algebras;
the relation between his « suitestypiques »
and our
homogeneous
verbal functions isexplained
in par. 8:roughly speaking,
some of his results can be obtained from oursby letting d
tend to
infinity.
2. - In this
paragraph
we consider the « binomial >>properties
ofsome
rings;
some of thefollowing elementary
remarks may be well-known,
but we do not know acomplete
source to refer to.Let I~ be an
integral
domain of characteristic zero. We shall denoteby
itsring
of fractions withinteger
denominators. If the element
as follows:
is defined
DEFINITION. A « binomial
ring »
is anintegral
domain C of char- acteristic zero, such that is in C for every c in C and every non-negative integer
k.Of
course, Z
is a binomialring (the
formulaaccounts for
negatives) . However,
if x is an indetermi-nate over
Z,
then thepolynomial ring Z[z]
is not binomial. Thereforewe introduce the set
(Polyals « ganzwertige Polynomen >>) .
It is well-known that consistsprecisely
of thosepolynomials
in which assumeinteger
values whenever x is
replaced by
aninteger.
This characterization has the immediate consequence is a binomialring.
Moregenerally,
if Z isreplaced by
any binomialring C,
we defineand
easily
check thatC f x}
is a binomialring. Clearly,
if u is an elementof a
ring containing C,
thenC f ~} _ C{x}}
is a minimalbinomial
ring containing
both C and u(thus
is the « binomialclosure >> of
C[x]).
If x2 , ... , xn arealgebraically independent
elementsover
C,
it is natural to defineby
inductionAn
element e
of thisring
can beexpressed
as a linear combinationI I , I I
with coefficients in C of
products I
where thekils
arenon-negative integers ;
e ischaracterized by the properties : ... , x~ ]
i... , cn) E C
for all
c,eC. In [1 ],
an element of7~~x~ , x2 , ... , x~,~
is called aninteger-producing polynomial
». Inparticular,
since2/}
=_ ~ f x~ {y~,
we must have « binomial identities » asfor some
LEMMA 1.
Any subring (with 1) o f
binomial.PROOF. Such a
ring
consists of all the fractionsrls
such that theprime
factors of sbelong
to a fixed set ofprimes. Then, by
the binomialidentities
above,
it suffices to prove that for anyprime p
we have1/p
thesubring generated by 1/p.
To thisaim
we writek ! = p-r with
p ~
r ; then1 /p
= a + rc for some a eZ,
c ehence, by
the binomialexpansion
forx + B
ky ,
/ it suffices to prove ? /when In fact we
compute
If d denotes a
positive integer,
we shall denoteby Zed]
thering
of the rational numbers
r/s
such that noprime dividing
s is =1 mod d(e.g. Z[2]
=~~21 ~ Z[4]
=1, t, ...]).
LEMMA 2. Zet oc be a
primitive (complex)
d-th rooto f unity. For
every
non-negative integer k
there exists apolynomial e,
e suchi
that k = k f or
all i c- 7 .PROOF. Let us write k ! =
ts,
where theprime
factors of t and sare =1 and
respectively #
1 mod d. In the factorring
the d-throots of
unity
areexactly d,
say where is a unit if mod d(for
t = p- this is in Lemma 1 of[4];
then the Chinese remainder theoremapplies).
In let usdivide and write
where Î’k
EZ[xl
is apolynomial
ofdegree
smaller than c.By
substitut-ing x
= ai we obtainSince we deduce
From this we shall deduce yk E and
as we wanted. In
fact,
let us denoteby y,,
theimage
of y,, under the naturalhomomorphism
of onto We know thatfor
i =1, 2, ... , d ;
thereforeyk
is apolynomial having
d distinct roots in anddegree
smaller thand ;
moreover, the difference of any two such roots is a unit inZ/tZ,
thusyk = 0
i. e. yk ELEMMA 3. =
PROOF.
By
Lemma1,
isbinomial; by
Lemma2,
contains(X
Bk
for every k. Since1/d
liesinz[dl
I we Asfor the inverse
inclusion,
we mustonly
prove or,equivalently,
whenever p is aprime, p - 1
mod d.Since this is trivial when p =
d,
we can assume p= dq --~- h,
1 C h d.Since all the elements of the field are roots of x, we have
Then =
-1 ) xh
we derive that the remainder/ uB
of the division
of I
Replacing x by
aayields
We now sum over i and take into account that I whenever 1. Then Lemma 2
yields
As a binomial
ring containing
oc, contains the left handterm,
thus also
lip,
sincep ,~
d.Let us notice that ai - mi is a unit in
whenever I fl j
mod d.This is
easily proved by letting r
= i- j
in thefollowing
relationLEMMA 4. Let
p be a prime integer, p ---1
mod d..Let m be any naturalenumber, a a primitive
d-th rootof
1 mod pm. Then there is aunique ring homomorphism of
ontomapping
Lx intoPROOF.
Any
element of Z[,] is written as a fractionrls,
withp ,~ s ;
therefore s is a unit
mod pm
and hence the naturalhomomorphism
of Z onto
Z/p-Z
isuniquely
extended to thering
of fractions Since isisomorphic
to the factorring
whereød
isthe
cyclotomic polynomial,
the statement will hold if we prove~d(a) ---
0mod pm.
From xd -1= TI ød,
we derive thatØd(a) =t=
0, d’jd
modpm implies ~~. (a) = 0 modpm’
for someSince divides
xd’-1,
thisimplies ad’
which is im-possible,
as a is aprimitive
d-th root of 1mod pm’ (see [4],
Lemma1).
Thus - 0
mod pm.
3. - We shall
operate
within a group whose existence and pro-perties
are thesubject
of a recent paperby
R. H. Bruck[1];
from itwe borrow definitions and
symbols,
and refer to it for details. Hereis the statement of Theor. 1.1 of
[1],
in aslightly simplified
form:Let
(A, )
be asimply
orderedset, consisting
of a finite number n > 0 ofelements; (A~, )
a universe of basicsymbols,
based onA;
Am
the subsetconsisting
of the elements oflength L(u)
= m.Let
(.F, ~ )
be a free group of rank n on a free set ofgenerators
the
corresponding
set of basiccommutators;
the subset of commutators whose total
weight equals
m.If C denotes a binomial
ring,
then there exists a group FO whose elements may beuniquely represented
as ordered infiniteproducts
, then for each we have
where ... uk are the elements of
A~
oflength
smaller thenL(u)
and Pu is auniquely
defined element of ... , xk; I i.e. aninteger-producing polynomial.
Fe is an extension of
.F’,
which is in factrepresented by
elements11 go-
with allexponents eu
in Z(as
in[2],
theor.11.2.4).
In Fe the «
c-th-power
» gl exists for everyg E F°,
c EC;
ifThe fact that
Pu
and Ru do notdep end
upon Cimplies
that fortheir determination one can assume C = Z and
operate
within the freenilpotent
group(here .Fm
denotes the m-th term of the lower central series of.F’) by
the«collecting
process »,Let us
point
out some conseguences of this theorem(not
allexplicitly
stated in[1]~ :
The c-th power
enjoys
the basicproperties
ofinteger
powers, i.e.the last formula
being
ageneralization
of P. Hall’s([2], 12.3.1).
By analogy
with thelanguage
ofC-modules,
one can introducethe
concepts
ofC-groups (i.e.
groups with c-th power for c EC), C-subgroups, C-homomorphisms
etc.by requiring
thecompatibility
with
taking
c-th powers. If one does so, some basicproperties
of Fare
naturally generalized
toFC;
forexample:
the lower central series of Fc consists ofC-subgroups,
each termF§’ consisting
of thoseelements
gl--
for which eu = 0 when thefactor-group
is a free
C-module;
sincenF’
=(1),
a statementholding
m
on may be
proved
to hold onFC, by
induction on m. Theseand others similar statements can be
proved by
a commonprocedure,
as follows: a
property
of F is r educed to a set of relations among the(integer) exponents
involved in theinfinite-product representation.
Because of their
polynomial
nature, the same relation must hold when Z isreplaced by C,
thusyielding
theanalogous property
for Fe.In par. 4 we shall exhibit some
examples
of thisprocedure.
Fc is called a «
completion
of F » withrespect
to thering
C. Here«
completeness »
meanssubstantially
that ifhl, h2,
... ,h~ , ...
is a sequence of elements of Fc such that forevery
then the sequence ofpartial products
converges to a
unique
element in otherwords k == kj modFf
for
every j.
Therefore some infiniteproducts
aremeaningful
inFO;
for
example,
the orderedproduct
is a well-defined element ofF, provided
thefollowing
condition holds: ifL(u)
= m, thenFc.
Infact,
the finiteness of Aimplies
thatonly
a finite number of basic commutators haveweight
m.4. - One of the basic
properties
of the free group Fimplies
thatfor any
integer
c there exists aunique endomorphism
of Fmapping
every free
generator ga (a
EA)
into its c-th power,The purpose of this
paragraph
is to extend in the natural way thisendomorphism
toh~, letting
c assume any value in C. To thisaim,
we start with the basic commutators and
define,
for anyc E C, v
EA.
LEMMA 5:
PROOF. Since
(c~)
is trueby
definition ifL(v)
=1,
we shall induceon m =
.L(v) .
Then we can assumeTherefore we calculate
(compare
par.3)
where
where
Similarly
we calculate the commutator and prove statement( a ) .
b )
It is a consequence of well-known commutator-relations([2],
18.4.10 andfoll. )
that in thegroup F
we have ---g:m
whenever
m = L(v).
This isequivalent
to8v,v(c)
= cm, = 0 forL(u) L(v)
orZ(~) = Z(v), ~ ~ v
for any cinteger.
isa
polynomial,
the same must hold when c assumes any value in thering
C.q.e.d.
Now let be any element of FC. We define
First of
all,
we must prove that there is noproblem
involved with the infinite number of factors.According
to the final remarks in par.3,
it will suffice to show that(gl,")e-
EFO
if m =L(v).
Infact
we claim
This follows from Lemma
5,
if we setLEMMA 6. Let
ments
of Aoo
whoselength
is smaller than m, andQu E
x1, ... ,PROOF. In order to calculate
fu
weonly
have to consider the finiteproduct
v. Since we havejust
seen that Ia
repeated
use of theproduct
formula(involving
thepolynomial Pu)
combined with Lemma 5
yields (b).
LEMMA 7. Let c E C. Then the
mapping g
r~C-endomorphism o f
FO.I f
thenPROOF. Let )e elements of po. We write
and use Lemma 6 to show that
where
But
when 9
E F and c EZ, by construction,
themapping g
r-~gici
induces an
endomorphism
of F. Therefore s,, = t~ in this case, hence alsoT u by
the familiarargument,
and(gg’)to
=The
proofs
that(gccl)e
e - and =g(cel
aresimilar,
involv-ing polynomials
in c, e, eUl’ ... ,We conclude this
paragraph
with a definition and some remarks which willsimplify
theproof
of our main theorem in par. 5.DEFINITION. A
mapping
E : will be called aC(z)-mapping
if for each u E
A.
there exists anelement Eu
E such thatfor every c E C .
For
example, if g
is a fixed element inFIO,
then the«exponential
function » c - gC is a
by
Lemma6,
the same holds forMore
generally,
ourprevious
remarks show thatproducts
and powers of
C{x}-mappings
are still such.5. - Let c be an element of the binomial
ring C, g -
the endo-morphism
of Fc of Lemma7, j
a natural number. The element will be defined to be «c-homogeneous
ofdegree j
» ifgc’
(i.e.
go is an «eigen-element » belonging
to the« eigenvalue » Ci).
Theterm is
suggested by
the fact that such an element willproduce
« c-homogeneous
functions » in the sense of par. 1.Let d be a
positive number,
and therings
defined in par.2;
weapply
the construction of par. 3 with theserings
in therole of C
(by
Lemma 1 and 4 both arebinomial)
and consider the groups Then F isnaturally
embedded inFZ[dJ
and thelatter in
The main result of this paper is
THEOREM 1.
Every
element g EFz Idi
can beuniquely
written asi.e. a
product o f a-homogeneous
elementsof degree j
=1, 2,
... , d.PROOF. Let us write for short = 1~’** = It may be worth to
point
outthat gi
is to be found in I’*although ga"’
isonly
defined in F**.By
Lemma7,
aa-homogeneous
element ofdegree j
is alsoai-homogeneous
ofdegree j
for allintegers
i.We shall
operate mod-F-**,
and induce on m.For m == 1 we define:
It follows from Lemma 5
(or directly
from the definitionof g
that any element of .F’** is
a-homogeneous
ofdegree 1, mod F’:*.
Therefore we have
the last congruence
being
trivialfor j .--- 2, ... , d.
The elements gl,;are in J~* and
they
areunique
sincefor j = ~, ... , d
thecondition
(gl,~)" = gi:~ --- (gl.~)"’ implies
thus- 1
Then the other condition
implies
also g1,1 -= g.Let us assume that we have
proved
the existence of d elements which areunique mod 0f/*
withrespect
to thefollowing
conditions
We must construct d elements E F*
satisfying
similar congruences and prove theirunicity
modI’m+1.
To thisaim,
we shallcompute
the «mean value » of for c
running
over the d-th roots ofunity.
Thus let us defineWe now consider the
mapping
of into F** definedby
Since lies in
F*,
the final remarks in par. 4 show that isa i.e.
By letting
c = (Xi(i = 1, 2, ... , d),
the inductiveassumption implies
~m,~(ai) ---1
therefore8,,(ai)
- 0 ifL(u)
m. Thus we haveIf we take into account that these ele- ments are central we can
compute:
We want to prove that the
exponent of g,~
lies inIn
fact, by
Lemma 2 for eachpositive integer k
there exists aai
polynomial
such it follows that forevery there is a
polynomial
such that Since for
d
I as we wanted. Therefore we have
proved For j =t= m mod d
let us define fromby simply cancelling
all factors ofweight greater
than m in its infinite-product representation; for y ==
m mod d let us defineyve must now prove the congruences We have
by
definitionHence
by
Lemma 7 wecompute
Here we have used the fact that for
all j
we haveso that
(mod .~’m+1)
the order of the factors in theproduct
isirrelevant,
and theexponent
(Xj may be taken out of theparenthesis.
We still have to consider the m mod d. First we notice that for m the construction above has
given
=then the inductive
assumption
g --- gm-1,1... gm-1~d combined with the definition of gm.3yields
also Since wehave
proved
the congruence we deduce that we can writewhere z,,,,
is aproduct
of powers of basiccommutators g.
ofweight L(u) =
~n. From Lemma 5 we know that for such commutators wehave
This
gives immediately
andfinally
As for
unicity,
let us assume thatsatisfy
the same
requirements
for -.By
the inductiveassumption
we canwrite,
Since the
previous argument gives
we cancompute
in two different ways,namely:
By comparison
weget ~2U~.,,,~~ai-"~ -1,
hence- 1 whenever j K m mod d.
Asfor 1 == m,
the first condition g -hm,1 ... hm,d implies wm,l . ·.’Zl~m,d ---1,
thus~,vm, f =1.
Thus the theorem iscompletely proved.
Let us remark that
along
theproof
the powermappings
g -g",
have both been used.
By
Lemma3,
is a minimalbinomial
ring containing
a andTherefore,
if thisproof
has tobe carried on in a group of
type
we cannot choose for C a smallerring
than For the same reason we cannotexpect
to find thehomogeneous
factors gj in a propersubgroup
of F* =FZcdH
evenif g lies in F.
6. The
proof
of theorem 1gives
a recursive method tocompute
gm,;
starting
with so that gj can be determined mod2~
forarbitrary
m. We shallapply
the firststeps
of thisprocedure
to theelement X Y of the free group F =
X, Y~ ,
whose elements are re-presented by
ordered infiniteproducts
asWe first choose d = 2
(d = 1
isuninteresting).
Then a = -1 and=
Zr2] = Z[’2-1-
Wecompute :
Let us now choose d = 3. We
compute
For d =
2, 3,
4 and withinweight
4(i.e.
mod.F’b )
we find the sameexponents
appearing
in[4],
p. 215. The reason for this coincidence(the computing
method was therefairly different)
will beexplained
in par. 7. From theor. 3 it will also be clear
why
two different valuesd d’
yield
the same elements forThe case d = 2 is
special:
if we start from aword g
E F we canavoid recursive formulas and basic commutators
expansions,
since the« mean value » of over the two roots ~1
gives directly
the exactformula for g1. In fact let us write
We claim:
f 2 = g2 .
Weeasily compute:
By unicity
we must havef 1=
911f
= g2 as we wanted.In
particular, if g
= X Y we find7. We shall
apply
theor. 1 to finite p-groups;possible general-
izations will be discussed later. In this
paragraph
P willalways
denotea finite p-group of
exponent pm, d
a divisor ofp -1, a
aprimitive
d-th root of
unity,
ymod pm.
There is a
unique
way toassign
to P a structure ofi.e. for any
h E P,
we define he to be theunique
element1~ E P such that ks = hr.
Equivalently,
we may consider the canonicalring-homomorphism
of Z ontoZ/p-Z,
extend it to thering
of fractionsZed] by letting
c =rls r-~ (r
+ + = b +pmZ
and define he = hb. P becomes also(although
notcanonically)
a ifwe define h" = ha i.e. we use the
homomorphism
of Lemma 4.We now choose n
generators
zi , x2 , ... , xn for P and denoteby F
the free group with free
generators
g1, g~, ... , gn . Then we see thatthere is a
unique
of _ .F’**Nonto
Psuch that
° ’
If v is such a
homomorphism,
it is clear how it must act on a basic commutator gu; as v is also then we must have for anywhere
only
finite factors are nontrivial,
as P isnilpotent.
Thisaccounts for
unicity.
As forexistence,
we maydefine gv by
the lastequation above;
theproof
that v is aZ[,,][a]-homomorphism
followseasily,
y once theegponents eu
have beenreplaced by
properintegers,
i.e.
by
thering homomorphism
of on we described in Lemma 4.As we have
already
seen in par.1,
thisimplies
inparticular
thatif (p
is anautomorphism
of P = z~ , ... ,Xn)
such that = xt" fori = 1, 2, ..., d then 99
is induced onby
the automor-phism 0:
9 F*9 (,x)
of .F’**.Now we reconsider the
homomorphism g 1--+ g
described in par.1, mapping
the free group I’ onto the groupP
of the verbal functions in n variables on the p-group P. Asbefore,
there is thepossibility
to extend this
mapping
to aZ[d][a]-homomorphism
of F**(the
finitep-group P
plays
now the role of Pabove):
iffor any
n-ple (hl,h2, ... , hn )
of elementsh i E P
we defineClearly,
ifgjllll
=g"’
in.F’**, then ij
isa-homogeneous
ofdegree j,
i.e.Then from Theor. 1 it follows
COROLLARY 1. Let P be a
f inite
p-groupof
exponent p-, d a divisorof p -1, a a primitive
d-th rootof
1mod pm.
Then every verbalf unc-
tion
(in n variables)
on P can beuniquely
written as aproduct of
da-homogeneo2cs
verbalfunctions of degree 1, 2,
... , d.PROOF.
Let 9
be a verbal function onP, corresponding
to theword We embed .F in .F’** and write g = gl g2 ... ga, I as in Theor 1. Then gj may be not in
.I’, but gi
is a verbal function on P(i.e.
there is aword 9, /
E.F’,
9although depending
onP,
such that91 = g3)
andclearly being a-homogeneous
ofdegree 1.
As for
unicity,
let us assume alsoliEF, ij a-homoge-
neous of
degree j.
We shall induce on the class c of P(or P),
thusassuming 1i == gi mod Pc .
Then if we set the element..., h~)
lies in the center of .P for any hence:Moreover,
theassumption ~1 ... ~a = gl ... 9a implies 21(h,
...,hn)
......
2,,(h,
7 ...,hn)
= 1. If wereplace
eachhj by
its powerh11
weget
for the
elements k, = z ~ ( hx , ... , hn )
thefollowing
relations:These are
equivalent
to a set of dhomogeneous
linearequations
inthe
ZlpmZ-module P,,,
whose coefficients have the(Vandermonde)
determinant Since ai - ai is a unit
mod pm,
there is«
only
the trivial solutions 1 =k2
= ... = kd = 1. Therefore = 1 and;j
as we wanted.When the
word-operation g
is the usual groupproduct,
associatedto the element
9 = XY
of the free group .F =(X, Y~ ,
then Cor. 1applies
to the situation we studied in[4],
andunicity
forcesj,
to bethe same
operation
which was there denotedby
the circle o; infact,
if 99
is anautomorphism
of P andx, y E P ;
thenyields
where the fact
that ii
is a verbal function has been used to write the firstequality.
Therefore the definition in[4]
was:x o y = gl(x, y).
This settles a
question
which arose in[4],
in connection with theindependence (upon P, a etc.)
of the rational numbers listed in the table of[4],
p.215; actually
theseexponents
must beindependent,
as
they
areuniquely
definedby
the factorizationof Theor.
1,
and this isuniquely
determinedby
the choice of d.Since this factorization is
possible
in but not inF,
it is alsoclear
why
theindependence
of the formulasrequires
the use of rationalnon-integer exponents.
Once this coincidence has been
recognized,
manyarguments
of[4]
may be
repeated
without substantialchanges ;
infact,
the automor-phism
(p is nolonger available,
but the firststeps
of theproof
inTheor. 1
imply
x 0 y --- xy modx, Y)2’
hence x o z = xz, x o(yz)
=(0153 0 y)
zwhenever z commutes with x, y ; on these relations most of the
proofs
in
[4]
are based. The result is thefollowing.
THEOREM 2. Let P be a
f inite
p-group.If
we choose a divisor dof p -1,
writeXY==glg2...gà
as in T heor. 1 anddefine
x o y ==
91(X, y) for
each x,y E P,
then we obtain aloop P, 0
with thefollowing properties:
a) P,
0 ispower- associative;
the orderof
an element is the same as in the group.b) If
the3-generators subgroups of
P havenilpotency
class c ethen
P,
0 iscentrally nilpotent of
class[(c
+d - 1)[d] ([...] = integer part of ...~ .
c) I f
a is a d-th rootof
1mod pm,
theexponent of P,
then thepower
mapping:
is aloop -automorphism of P,
o.Here are some
particular
cases: if d =2,
we find x 0 y = theloop
of Glauberman(see
ref. in[4]) ;
if d = p - 1 and c C p,then
P,
0 is an abelian group(as
we shall see, this is the additive groupof the
Lie-algebra
associated toP, according
to Lazard[3]).
Sincex - y E
x, y~,
anysubgroup
of P is asubloop
ofP, o ,but
the converseis not
true;
forexample,
the setPi
of the «eigenelements
» for anautomorphism
of order d(see
par.1 )
is asubloop
but need not bea
subgroup.
Since x o y is a verbaloperation
onP,
agroup-auto- morphism
of P is aloop-automorphism
ofP, o; again,
the converseis not
true,
themapping being
acounterexample.
Thefollowing application
may deserve some attention: if the3-generators subgroups
of the p-group P have class p, then the group of the
automorphisms
of P is
isomorphically
embedded in the group of theautomorphisms
of the abelian group
P,
o.The
assumptions
of Cor. 1 can beweakened,
but some cautionis
needed;
forexample,
we may remark that the finiteness of P wasonly
used to ensure both the finiteexponent
and thenilpotency
of P.The former is not
required
to define a structure of on ap-group, since the
ring Z/p-Z
can bereplaced by
thering
ofp-adic integers;
but in this case, if g, E Igj OF,
thengj
may not bea verbal function on P.
However,
a more substantial obstacle togeneralizations
is met if one wants to omit thenilpotency assumption
for
P,
as thehomomorphism g - )
is definedonly
if the infinite pro- ducts involved converge in P. On the otherhand,
there is nodifficulty
in
formulating
Cor. 1 for any finitenilpotent
group such that p ---1 mod d for everyprime p dividing
the order of P.8. In this
paragraph
we shallexplain
the connection between thepreceeding
theorems and the results of M. Lazard([3]).
THEOREM 3. Let C be a binomial
ring eontain2ng let g be
a-rzelement
of
= gl g2 ... gd thefactorization of
Theor 1. Thenfor
every cc C we have
We
only give
an outline of theproof,
whose details wouldrequire
first to prove an
improvement
of Theor. 1.1 of[1 ],
in order to takeinto account the
degrees
of thepolynomials Pu,
Ru(see
par.2)
inthe
single
variables. One finds that theCfx}-mapping
of C into pc definedby
is such that the
degree of gu
in x does not exceedL(~c) .
On the otherhand,
the conditions(g,)0"
for i= 1, 2, ... , d implies
that eache. vanishes for x = a,
oe 2, == 1;
sinceclearly e
vanishes also for the(d+1)-th
value one deduces~u = 0
wheneverThen
(gicJ)-C
=1 as we wanted.For j = 2,
... , d theproof
is similar.
COROLLARY 2. Let P be a p-group whose
n-generators subgroups
have
nilpotency
class notexceeding
d. Then every verbalf unction
in nvariables on P can be
uniquely
written I wheregj
ise-homogerceous of degree j for
everyinteger
c.The
proof
isimmediate, by
Theor. 3 and the usualhomomorphism.
There is a connection between these statements and some results of M. Lazard
([3]);
he introduced theconcept
of « suitetipique >>
andproved
that such a sequence can beuniquely represented
as an infiniteproduct ([3],
p.143)
where t is a
parameter assuming non-negative integer
valuesand b~
lies in a suitable
extension,
sayE,
of the free group F.Roughly speaking,
this is the factorization of our Theor. 3for g
=g(1 )
if welet d tend to
infinity,
thusZldl
tend to the field ofrationals Q
andF* =
FZcdJ to FQ, in
the role of E. In thefollowing example
thisconnection will appear more
precise:
letas in
[3],
theor. 2.4. Now let usapply
Theor. 3 to the word g = X Y ==~1~2...~. We have f or
every c. In
particular,
if we let c assumeinteger
values t and embedF* in E we obtain
By comparison
it iseasily
seen that gi -bl,
g2 ---b2,
..., gd ---Likewise,
if .P is a p-group of class notexceeding
p -1,
then ourCor. 2
applied
tog = X Y
ford = p -1 yields substantially
Lem-ma 4.5 of
[3]; by letting
we
assign
to P a structure ofLie-algebra
from which theoriginal group-multiplication
is obtainedthrough
the Hausdorff formula. Thefollowing (open) questions naturally
arise:Under the
assumptions
of Cor. 1(i.e.
without any bound for the class ofP)
let us still writeg = XY, gl (x, y ) = x -E- y, (g(x, y ) ) 2 =
-
y~.
ThenP,
+ is aloop (Theorem 2),
but what kind of structure is(P, -~--, Q, ~) ~
Does there exist a formulaenabling
one to reconstruct the groupmultiplication starting
with the two newoperations
REFERENCES
[1] R. H. BRUCK: Completion of free groups by means of infinite product representation, Math. Zeit., 118 (1970), 9-29.
[2] M. HALL jr.: The
Theory
ofGroups,
Mac Millan Co., New York, 1959.[3] M. LAZARD: Sur les groupes nilpotents et les anneaux de Lie, Ann. E.N.S.,
71 (1954), 101-190.
[4] B. SCIMEMI: p-groups, diagonalizable automorphisms and
loops,
Rend. Sem.Mat. Padova, 45 (1971), 199-221.
Manoscritto pervenuto in redazione il 9 novembre 1972.